Abstract:
In an era where virtually every electronic device is built with the capability to connect
to the internet, and thus to the cloud, end users now are getting more adjoined with
numerous IoT devices which are part of a myriad of Smart Systems. Moreover, these
heterogeneous IoT environments, with newly boosted security features - thanks to
exponential growth in security solutions for IoT devices - possess unique behavioral
patterns in unique environments. Additionally, it is highly unlikely that in such
an environment, with plethora of device-types,every device will leak information at
once. Thus, in this work, we propose a security framework to establish secured
communication between end-user and the cloud using the behavioral patterns of the
IoT devices which are accessed by both the communicating parties following some
proper authorization. To implement our proposal, we have used Sensorscope sensor
network's weather sensor data. After training an Long Short Term Memory network
model using time series of sensor data, we predicted session keys between the cloud
and the user using noisy data. The goals we attained from this work are Twofold.
First, we achieved forward secrecy using session keys which are generated using noisy
environment data. Second, it is observed that when we decrypted the messages using
noisy-data-generated session keys, the accuracy of decryption varied according to the
proportion of the added noise in the sensor data. For keys generated with normalized
noise with 3% standard deviation, we found out decryption accuracy to be as high
as 96%. On the other hand, communicating parties from two di erent environments
can only decrypt only 50% of the message bits accurately. Finally we argued, since
the noise in sensor data is re
ected in the decryption accuracy, successful decryption
of messages with narrow margin of error veri es that the communicating parties are
part of the same environment and thus any intruder with information of the partial
environment cannot communicate without decryption accuracy falling drastically.
